The unit of analysis in the logistic regression model” is the individual birthing by a mother. We distinguish two kinds of outcome:

• Maternal death, with the outcome scored 1 for death within sixty days, else 0.

• Background death, with the outcome scored 1 for death within 61—730 days, else 0, but women who died within sixty days were excluded.

Independent variables are, depending on the model:

• ⅞^e∙

• Parity (dummies, grouped with parity group 2—4 omitted since it has expectably the lowest mortality: 1, 5—7, 8+, coded 1 if the parity of the birth falls in any group, else 0).

• Parish (dummies, with Cernik, the civil parish, omitted, coded 1 if the birth was in each parish, else 0).

• The preceding birth interval, or for parity 1, the interval from marriage.

• The mean lifetime birth interval.

• Crisis period (the occurrence of any of the crises described above, thus coded 1 for a crisis period, else 0). Multi-year crises were specified to include all of the relevant years. Single-year crises were coded to include the year of the crisis and the following year, since mobilization late in a year would imply absence of males early in the next.

• Multiple birth (coded 1 if the birth was a multiple birth, else 0).

• Year (calendar year of the birth minus 1815). Effectively, this is the influence of calendar year on higher order births; see below.

• Parity1*year (interaction of Parity 1 and Year, thus the effect of calendar year on parity 1 births only).

• BM (sometimes the variable of interest, also a covariate where MM is the dependent variable). Using this as a predictor effectively turns analysis into that of net MM.³²

Table 11.3 shows the resulting estimates and corresponding probability values.³³ Because we have prior reason to anticipate the direction of effect of each of the variables, from theory or from broader data, we accept as significant those results for which the two-tailed probability is less than 0.10, thus the one-tailed alpha level is 0.05.

In Table 11.3 are three sets of columns. The first lists the covariates. The second gives results for MM (0—60 days), and the third for BM (61—730 days). Readers should use Table 11.3 to make their own judgements about the direction and significance of effect of variables.

Table 11.3 Results of logistic regression

y>of MM is of particular interest.

That the physiological variables of age and parity show generally expectable patterns with respect to MM and BM gives us further confidence in the reliability of the data.

The long-term increase of MM and BM parallels some specific, and some general indicators of social and economic stress. The elasticities of mortality and fertility with respect to grain prices, hence the vulnerability of the population to short-term food crises, show something of the same trend, although it is raggedly interrupted after 1850 and shows differences between civil and military parishes. Evidence of nutritional insufficiency in the declining stature of military recruits in neighbouring Hungary follows a parallel trend but only up to 1850. Infant mortality shows the same increasing trend as MM and BM, possibly independently, but possibly affected directly by the deaths of mothers, inability to lactate, and losses of household economies of scale for surrogate nursing. At a more general level, there were changes in social and economic structures. The feudal system collapsed; whatever its burdens, its customary protections also collapsed, and for civil serfs there was an added burden of amortization. Military obligations lessened after the Napoleonic wars, and there were few significant effects of mobilizations on MM and BM thereafter; nevertheless MM and BM did increase after 1815. The need for cash increased over time, especially as feudal dues were increasingly converted to monetary taxes and as amortization payments had to be met.

The episodic peaks in MM are of particular interest. Graphical presentation shows a reasonable match between such peaks and periods of military mobilization (see Figure 11.5). The regressions show that MM is positively correlated with such mobilizations in general, although not always with any particular mobilization. That coincidence is not apparent for BM. We would expect that exogenous disease

effects would become more important after the sixty-day cut-off for MM, while direct obstetrical deaths would predominate in the first day or so of the sixty-day span, with puerperal disease emerging thereafter, then deaths from other causes but exacerbated by parturition. Since the general crisis effect and many of the individual crises effects are strong for MM but not for BM, we infer that the effect of crisis was felt predominantly on factors directly related to parturition, and that disease introduced by soldiers returning home was of lesser consequence. Since it was the absence of males, rather than their presence that seems operative, our attention focuses on the withdrawal of male labour from the household, on the shift of labour burden to women, some of whom were pregnant, parturient, or nursing, others of whom would otherwise have been able to assume the ordinary labour burdens of the former. Losses of household economies of scale as the joint family system decayed would have exacerbated these matters. The secular trend towards wage labour, drawing men away from the family farm, would have increased these effects over time.³⁴

That women at parity 1 were at higher risk may not be entirely a physiological phenomenon but may reflect the inferior position and greater labour demands on new brides, as well as the juniority and probably more frequent absence of their husbands. It is noteworthy that the secular increase in mortality occurs at a higher rate at parity 1 than at later parities. Further, the episodic increase at times of military mobilization is generally greater at parity 1 than at later parities.

8.